Statistically, some working mothers will answer the phone and finish the survey. If the survey was done correctly, and sampling bias and all that is accounted for, 1800 respondents is plenty to get a good representative picture of Americans. Surveys can work very well; even when using non-ideal survey methods.
Anecdotally, this survey seems to align pretty well with the people I know. E.g. people complaining about a pie costing something like 30% more than 5 years ago, while their income doubled or tripled in that same timeframe. Help-wanted signs in my area seem to be advertising wages around 50% more than 2019 as well.
LLMs do sometimes hallucinate even when giving summaries. I.e. they put things in the summaries that were not in the source material. Bing did this often the last time I tried it. In my experience, LLMs seem to do very poorly when their context is large (e.g. when "reading" large or multiple articles). With ChatGPT, it's output seems more likely to be factually correct when it just generates "facts" from it's model instead of "browsing" and adding articles to its context.